Turnover covariate data prep

library(data.table)
library(ggplot2)

Registered S3 method overwritten by 'dplyr':
  method           from
  print.rowwise_df     

library(beanplot) # for beanplots
library(gridExtra) # to combine ggplots together
library(grid) # to combine ggplots together


knitr::opts_knit$set(root.dir = rprojroot::find_rstudio_root_file()) # tell RStudio to use project root directory as the root for this notebook. Needed since we are storing code in a separate directory.

Load data

# BioTime
load('data/biotime_blowes/bt_malin.Rdata')
bt <- data.table(bt_malin); rm(bt_malin)

# Temperature average, trends, and seasonality
temperature <- fread('output/temperature_byrarefyID.csv.gz')

# microclimates
microclim <- fread('output/microclimates.csv.gz', drop = 1)

# NPP
npp <- fread('output/npplandocean.csv.gz')

# Body size
bs <- fread('output/mass_byrarefyid.csv.gz', drop = 1)
bs[, ':='(STUDY_ID = NULL, REALM = NULL, taxa_mod = NULL)] # remove unnecessary columns 

# Mobility
speed <- fread('output/speed_byrarefyID.csv.gz', drop = 1)
speed[, ':='(STUDY_ID = NULL, REALM = NULL, taxa_mod = NULL)] # remove unnecessary columns 

# Lifespan
lsp <- fread('output/lifespan_byrarefyID.csv.gz')

# CTI
cti <- fread('output/cti_byrarefyID.csv.gz')
    
# consumer vs. producer
consfrac <- fread('output/consfrac_byrarefyID.csv.gz')

# richness
rich <- fread('output/richness_by_rarefyID.csv.gz') # number of species

# endotherm vs. ectotherm
endofrac <- fread('output/endofrac_byrarefyID.csv.gz') # endotherm vs. ectotherm classifications

# human impact
human <- fread('output/humanimpact_by_rarefyID.csv.gz')

# %veg
veg <- as.data.table(readRDS('output/vct_by_rarefyID.rds'))
veg[, veg := (`tree cover % (mean)` + 0.5 * `non-tree veg. % (mean)`)/100] # veg index from 0 (all non-veg) to 1 (all tree). Non-tree veg counts as 0.5.

Plot a turnover example

First is a long time-series, next one shows an example result of removing the first year of self-comparison.

#bt[, .(n = .N), by = rarefyID][n > 50,]

ggplot(bt[rarefyID == '339_1085477'], aes(YEAR, Jtu_base)) +
    geom_point() +
    geom_smooth(method = 'lm') +
    xlab('Year') + ylab('Jaccard dissimilarity')

ggplot(bt[rarefyID == '489_16754'], aes(YEAR, Jtu_base)) +
    geom_point() +
    geom_smooth(method = 'lm', se = FALSE) +
    xlab('Year') + ylab('Jaccard dissimilarity') +
  geom_smooth(data = bt[rarefyID == '489_16754' & YEAR != 2000], method = 'lm', se = FALSE, color = 'red')

Assemble dataset of beta diversity trends (temporal turnover) and covariates

trends7 <- bt[, calctrendrem0(1-Horn_base, YEAR, 'Horntrendrem0'), 
    by = .(rarefyID)]

essentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliableessentially perfect fit: summary may be unreliable

Add covariates

# add covariates
trends <- merge(trends, temperature[, .(rarefyID, tempave, tempave_metab, temptrend, seas)], all.x = TRUE, by = 'rarefyID') # temperature ave, ave metabolic, trend, and seasonality
trends <- merge(trends, microclim[, .(rarefyID, microclim = Temp_sd20km)], all.x = TRUE, by = 'rarefyID') # microclimates
trends <- merge(trends, npp, all.x = TRUE, by = 'rarefyID') # npp
trends <- merge(trends, bs[, .(rarefyID, mass_mean_weight, mass_sd_weight)], all.x = TRUE) # body size mass (g)
trends <- merge(trends, speed[, .(rarefyID, speed_mean_weight, speed_sd_weight)], all.x = TRUE) # speed (km/hr)
trends <- merge(trends, lsp[, .(rarefyID, lifespan_mean_weight, lifespan_sd_weight)], all.x = TRUE) # lifespan (yr)
trends <- merge(trends, cti[, .(rarefyID, thermal_bias)], all.x = TRUE) # thermal bias (degC)
trends <- merge(trends, consfrac, all.x = TRUE) # fraction consumers
trends <- merge(trends, rich, all.x = TRUE) # species richness
trends <- merge(trends, endofrac, all.x = TRUE) # endotherm vs. ectotherm
trends <- merge(trends, human[, .(rarefyID, human_bowler = atc, human_venter = hfp, human_halpern = himp)], all.x = TRUE) # human impact
trends <- merge(trends, veg[, .(rarefyID, veg = veg)], all.x = TRUE) # vegetation index
trends[REALM == 'Marine', veg := 0] # veg index is 0 at sea

Do some basic checks of the turnover calculations

# basic checks
trends

trends[, .(minJtu = min(Jtutrend), maxJtu = max(Jtutrend), minJbe = min(Jbetatrend), maxJbe = max(Jbetatrend), 
           minHo = min(Horntrend, na.rm = TRUE), maxHo = max(Horntrend, na.rm = TRUE)), by = REALM]

trends[, .(nJtu = sum(Jtutrend < 0)), by = REALM] # why are some turnover trends < 0? first year more diverged than future years

trends[, .(nJ = sum(Jtutrend > 0)), by = REALM]

# number of species
trends[, summary(Nspp)]

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
   1.00    6.00   11.00   19.35   24.00 1427.00 

trends[, plot(Nspp, Jtutrend, log = 'x')]

NULL

Remove studies with only 1 species

nrow(trends)

[1] 53467

trends <- trends[Nspp > 1, ]
nrow(trends)

[1] 53013

Turnover calculations are correlated, though less so for Horn

# are turnover calculations correlated?
ggplot(trends, aes(Jbetatrend, Jtutrend)) +
    geom_point(alpha = 0.3) +
    geom_smooth()

ggplot(trends, aes(Jbetatrend, Horntrend)) +
    geom_point(alpha = 0.3) +
    geom_smooth()

Temporal turnover is not all that correlated between including first year or not.

ggplot(trends, aes(Jtutrend, Jtutrendrem0)) +
  geom_point(alpha = 0.3) +
  geom_smooth() +
  geom_abline(a = 0, b = 1)

Ignoring unknown parameters: a, b

ggplot(trends, aes(Jbetatrend, Jbetatrendrem0)) +
  geom_point(alpha = 0.3) +
  geom_smooth() +
  geom_abline(a = 0, b = 1)

Ignoring unknown parameters: a, b

ggplot(trends, aes(Horntrend, Horntrendrem0)) +
  geom_point(alpha = 0.3) +
  geom_smooth() +
  geom_abline(a = 0, b = 1)

Ignoring unknown parameters: a, b

Compare covariates across realms

i <- trends[, !duplicated(rarefyID)]; sum(i)

[1] 53013

par(mfrow=c(5,3))
beanplot(rarefyID_y ~ REALM, data = trends[i,], what = c(1,1,1,1), col = c("#CAB2D6", "#33A02C", "#B2DF8A"), border = "#CAB2D6", ylab = 'Latitude (degN)', ll = 0.05)
beanplot(tempave ~ REALM, data = trends[i,], what = c(1,1,1,1), col = c("#CAB2D6", "#33A02C", "#B2DF8A"), border = "#CAB2D6", ylab = 'Temperature (degC)', ll = 0.05)

beanplot(tempave_metab ~ REALM, data = trends[i,], what = c(1,1,1,1), col = c("#CAB2D6", "#33A02C", "#B2DF8A"), border = "#CAB2D6", ylab = 'Metabolic Temperature (degC)', ll = 0.05, bw = 'nrd0') # nrd0 bandwidth to calculation gap
beanplot(seas ~ REALM, data = trends[i,], what = c(1,1,1,1), col = c("#CAB2D6", "#33A02C", "#B2DF8A"), border = "#CAB2D6", ylab = 'Seasonality (degC)', ll = 0.05)

beanplot(microclim ~ REALM, data = trends[i,], what = c(1,1,1,1), col = c("#CAB2D6", "#33A02C", "#B2DF8A"), border = "#CAB2D6", ylab = 'Microclimates (degC)', ll = 0.05)

log="y" selected

beanplot(temptrend ~ REALM, data = trends[i,], what = c(1,1,1,1), col = c("#CAB2D6", "#33A02C", "#B2DF8A"), border = "#CAB2D6", ylab = 'Temperature trend (degC/yr)', ll = 0.05)

beanplot(mass_mean_weight ~ REALM, data = trends[i,], what = c(1,1,1,1), col = c("#CAB2D6", "#33A02C", "#B2DF8A"), border = "#CAB2D6", ylab = 'Mass (g)', ll = 0.05, log = 'y')
beanplot(speed_mean_weight +1 ~ REALM, data = trends[i,], what = c(1,1,1,1), col = c("#CAB2D6", "#33A02C", "#B2DF8A"), border = "#CAB2D6", ylab = 'Speed (km/hr)', ll = 0.05, log = 'y')

beanplot(lifespan_mean_weight ~ REALM, data = trends[i,], what = c(1,1,1,1), col = c("#CAB2D6", "#33A02C", "#B2DF8A"), border = "#CAB2D6", ylab = 'Lifespan (yr)', ll = 0.05, log = 'y')
#beanplot(consfrac ~ REALM, data = trends[i,], what = c(1,1,1,1), col = c("#CAB2D6", "#33A02C", "#B2DF8A"), border = "#CAB2D6", ylab = 'Consumers (fraction)', ll = 0.05, log = '') # too sparse
#beanplot(endofrac ~ REALM, data = trends[i,], what = c(1,1,1,1), col = c("#CAB2D6", "#33A02C", "#B2DF8A"), border = "#CAB2D6", ylab = 'Endotherms (fraction)', ll = 0.05, log = '') # too sparse
beanplot(Nspp ~ REALM, data = trends[i,], what = c(1,1,1,1), col = c("#CAB2D6", "#33A02C", "#B2DF8A"), border = "#CAB2D6", ylab = 'Number of species', ll = 0.05, log = 'y')

beanplot(thermal_bias ~ REALM, data = trends[i & !is.na(thermal_bias),], what = c(1,1,1,1), col = c("#CAB2D6", "#33A02C", "#B2DF8A"), border = "#CAB2D6", ylab = 'Thermal bias (degC)', ll = 0.05)
beanplot(npp ~ REALM, data = trends[i,], what = c(1,1,1,1), col = c("#CAB2D6", "#33A02C", "#B2DF8A"), border = "#CAB2D6", ylab = 'NPP', ll = 0.05)

log="y" selected

beanplot(veg ~ REALM, data = trends[i & REALM !='Marine',], what = c(1,1,1,1), col = c("#CAB2D6", "#33A02C", "#B2DF8A"), border = "#CAB2D6", ylab = 'NPP', ll = 0.05)

Marine are in generally warmer locations (seawater doesn’t freeze) Marine have much lower seasonality. Marine and freshwater have some very small masses (plankton), but much of dataset is similar to terrestrial. Marine has a lot of slow, crawling organisms, but land has plants. Land also has birds (fast).

Plot turnover vs. temperature

Time-series length and temperature trend?

ggplot(trends, aes(temptrend, nyrBT)) +
  geom_point() +
  geom_smooth() +
  scale_y_log10()

Plot turnover vs. explanatory variables

Lines are ggplot smoother fits.

Strong trends with temperature change, but trends are pretty symmetric around no trend in temperature, which implies warming or cooling drives similar degree of community turnover. Some indication of less turnover for larger organisms (mass) Higher turnover on land with higher seasonality? More turnover for shorter-lived organisms? No really clear differences among realms.

Write out

write.csv(trends, gzfile('output/turnover_w_covariates.csv.gz'), row.names = FALSE)

Useful variables

# realm that combined Terrestrial and Freshwater, for interacting with human impact
trends[, REALM2 := REALM]
levels(trends$REALM2) = list(TerrFresh = "Freshwater", TerrFresh = "Terrestrial", Marine = "Marine")

# group Marine invertebrates/plants in with All
trends[, taxa_mod2 := taxa_mod]
trends[taxa_mod == 'Marine invertebrates/plants', taxa_mod2 := 'All']

Log-transform some variables, then center and scale.

trends[, tempave.sc := scale(tempave)]
trends[, tempave_metab.sc := scale(tempave_metab)]
trends[, seas.sc := scale(seas)]
trends[, microclim.sc := scale(log(microclim))]
trends[, temptrend.sc := scale(temptrend, center = FALSE)]
trends[, temptrend_abs.sc := scale(abs(temptrend), center = FALSE)] # do not center, so that 0 is still 0 temperature change
trends[, mass.sc := scale(log(mass_mean_weight))]
trends[, speed.sc := scale(log(speed_mean_weight+1))]
trends[, lifespan.sc := scale(log(lifespan_mean_weight))]
trends[, consumerfrac.sc := scale(consfrac)]
trends[, endothermfrac.sc := scale(endofrac)]
trends[, nspp.sc := scale(log(Nspp))]
trends[, thermal_bias.sc := scale(thermal_bias)]
trends[, npp.sc := scale(log(npp))]
trends[, veg.sc := scale(log(veg+1))]
trends[, human_bowler.sc := scale(log(human_bowler+1)), by = REALM2] # separate scaling by realm
trends[REALM2 == 'TerrFresh', human_footprint.sc := scale(log(human_venter+1))]
trends[REALM2 == 'Marine', human_footprint.sc := scale(log(human_halpern))]

Do the variables look ok?

Unscaled

# histograms to examine
cexmain = 0.6
par(mfrow = c(5,4))
invisible(trends[, hist(minyrBT, main = 'Start year', cex.main = cexmain)])
invisible(trends[, hist(maxyrBT - minyrBT, main = 'Duration (years)', cex.main = cexmain)])
invisible(trends[, hist(nyrBT, main = 'Number of sampled years', cex.main = cexmain)])
invisible(trends[, hist(mass_mean_weight, main = 'Mass (g)', cex.main = cexmain)])
invisible(trends[, hist(speed_mean_weight, main = 'Speed (km/hr)', cex.main = cexmain)])
invisible(trends[, hist(lifespan_mean_weight, main = 'Lifespan (yr)', cex.main = cexmain)])
invisible(trends[, hist(tempave_metab, main = 'Metabolic temperature (°C)', cex.main = cexmain)])
invisible(trends[, hist(consfrac, main = 'Consumers (fraction)', cex.main = cexmain)])
invisible(trends[, hist(endofrac, main = 'Endotherms (fraction)', cex.main = cexmain)])
invisible(trends[, hist(tempave, main = 'Environmental temperature (°C)', cex.main = cexmain)])
invisible(trends[, hist(temptrend, main = 'Temperature trend (°C/yr)', cex.main = cexmain)])
invisible(trends[, hist(seas, main = 'Seasonality (°C)', cex.main = cexmain)])
invisible(trends[, hist(microclim, main = 'Microclimates (°C)', cex.main = cexmain)])
invisible(trends[, hist(Nspp, main = 'Species richness', cex.main = cexmain)])
invisible(trends[, hist(thermal_bias, main = 'Thermal bias (°C)', cex.main = cexmain)])
invisible(trends[, hist(npp, main = 'Net primary productivity', cex.main = cexmain)])
invisible(trends[, hist(veg, main = 'Vegetation index', cex.main = cexmain)])
invisible(trends[, hist(human_bowler, main = 'Human impact score (Bowler)', cex.main = cexmain)])
invisible(trends[, hist(human_venter, main = 'Human impact score (Venter)', cex.main = cexmain)])
invisible(trends[, hist(human_halpern, main = 'Human impact score (Halpern)', cex.main = cexmain)])

Scaled

# histograms to examine
cexmain = 0.6
par(mfrow = c(5,4))
invisible(trends[, hist(tempave.sc, main = 'Environmental temperature (°C)', cex.main = cexmain)])
invisible(trends[, hist(tempave_metab.sc, main = 'Metabolic temperature (°C)', cex.main = cexmain)])
invisible(trends[, hist(seas.sc, main = 'Seasonality (°C)', cex.main = cexmain)])
invisible(trends[, hist(microclim.sc, main = 'log Microclimates (°C)', cex.main = cexmain)])
invisible(trends[, hist(temptrend.sc, main = 'Temperature trend (°C/yr)', cex.main = cexmain)])
invisible(trends[, hist(temptrend_abs.sc, main = 'abs(Temperature trend) (°C/yr)', cex.main = cexmain)])
invisible(trends[, hist(mass.sc, main = 'log Mass (g)', cex.main = cexmain)])
invisible(trends[, hist(speed.sc, main = 'log Speed (km/hr)', cex.main = cexmain)])
invisible(trends[, hist(lifespan.sc, main = 'log Lifespan (yr)', cex.main = cexmain)])
invisible(trends[, hist(consumerfrac.sc, main = 'Consumers (fraction)', cex.main = cexmain)])
invisible(trends[, hist(endothermfrac.sc, main = 'Endotherms (fraction)', cex.main = cexmain)])
invisible(trends[, hist(nspp.sc, main = 'log Species richness', cex.main = cexmain)])
invisible(trends[, hist(thermal_bias.sc, main = 'Thermal bias (°C)', cex.main = cexmain)])
invisible(trends[, hist(npp.sc, main = 'log Net primary productivity', cex.main = cexmain)])
invisible(trends[, hist(veg.sc, main = 'log Vegetation index', cex.main = cexmain)])
invisible(trends[, hist(human_bowler.sc, main = 'log Human impact score (Bowler)', cex.main = cexmain)])
invisible(trends[, hist(human_footprint.sc, main = 'log Human impact score (Venter & Halpern)', cex.main = cexmain)])

Check correlations among variables. Pearson’s r is on the lower diagonal.

panel.cor <- function(x, y, digits = 2, prefix = "", cex.cor, ...)
{
    usr <- par("usr"); on.exit(par(usr))
    par(usr = c(0, 1, 0, 1))
    r <- cor(x, y, use = 'pairwise.complete.obs')
    txt <- format(c(r, 0.123456789), digits = digits)[1]
    txt <- paste0(prefix, txt)
    if(missing(cex.cor)) cex.cor <- 0.8/strwidth(txt)
    text(0.5, 0.5, txt) #, cex = cex.cor * r)
}
pairs(formula = ~ tempave.sc + tempave_metab.sc + seas.sc + microclim.sc + temptrend.sc + temptrend_abs.sc + mass.sc + speed.sc + lifespan.sc + consumerfrac.sc + endothermfrac.sc + nspp.sc + thermal_bias.sc + npp.sc + veg.sc + human_bowler.sc + human_footprint.sc, data = trends, gap = 1/10, cex = 0.2, col = '#00000022', lower.panel = panel.cor)

Mass and lifespan look tightly correlated, but r only 0.56…? Tempave_metab and lifespan don’t look tightly correlated, but r= -0.81 Tempave_metab and speed don’t look tightly correlated, but r= -0.83 Lifespan and speed don’t look tightly correlated, but r = 0.73